Rethinking Artistic Copyright Infringements in the Era of Text-to-Image Generative Models (2404.08030v1)

Abstract: Recent text-to-image generative models such as Stable Diffusion are extremely adept at mimicking and generating copyrighted content, raising concerns amongst artists that their unique styles may be improperly copied. Understanding how generative models copy "artistic style" is more complex than duplicating a single image, as style is comprised by a set of elements (or signature) that frequently co-occurs across a body of work, where each individual work may vary significantly. In our paper, we first reformulate the problem of "artistic copyright infringement" to a classification problem over image sets, instead of probing image-wise similarities. We then introduce ArtSavant, a practical (i.e., efficient and easy to understand) tool to (i) determine the unique style of an artist by comparing it to a reference dataset of works from 372 artists curated from WikiArt, and (ii) recognize if the identified style reappears in generated images. We leverage two complementary methods to perform artistic style classification over image sets, includingTagMatch, which is a novel inherently interpretable and attributable method, making it more suitable for broader use by non-technical stake holders (artists, lawyers, judges, etc). Leveraging ArtSavant, we then perform a large-scale empirical study to provide quantitative insight on the prevalence of artistic style copying across 3 popular text-to-image generative models. Namely, amongst a dataset of prolific artists (including many famous ones), only 20% of them appear to have their styles be at a risk of copying via simple prompting of today's popular text-to-image generative models.

• The paper introduces ArtSavant, a novel framework that quantitatively evaluates artistic style copying to uncover potential copyright infringements.
• It employs a dual-method approach with DeepMatch achieving 89.3% accuracy and an interpretable TagMatch for nuanced style classification.
• The study finds that about 20% of artists' styles are at risk, underscoring the need for policy adaptation in the era of AI-driven art.

Rethinking Artistic Copyright Infringements with ArtSavant: Insights on Style Copying by Generative Models

Introduction

The advent of text-to-image generative models has brought about a new set of challenges in copyright law, notably in terms of artistic style infringement. These generative models, including Stable Diffusion and Imagen, have shown proficiency in mimicking stylistic elements peculiar to individual artists, hence posing potential copyright issues. This paper introduces ArtSavant, a tool designed for identifying unique artistic styles and determining if such styles have been reused in generated images. The focus lies on articulating the notion of "artistic style" beyond mere image-wise similarity, proposing a classification over sets of images to quantitatively assess style copying.

Artistic Style Classification

The concept of "artistic style" is complex, often involving elements or signatures that recur across an artist's body of work. The problem's formulation transitions from individual image analysis to evaluating sets of images to identify style replication effectively. ArtSavant operates on a curated dataset from WikiArt, comprising works from 372 artists, leveraging two distinct approaches for style classification: DeepMatch and TagMatch. DeepMatch employs a neural architecture for artist classification with an aggregation scheme based on majority voting. TagMatch, in contrast, introduces an interpretable, tag-based classification, enabling a deeper analysis of unique stylistic elements.

Methodology

• DeepMatch Strategy: This component of ArtSavant classifies artworks to their corresponding artists by training on embeddings generated from a CLIP ViT-B/16 vision encoder. The model achieves considerable success in distinguishing artists based on their latent style signatures, with an overall accuracy of 89.3% on the test dataset. This finding substantiates the existence of distinctive artistic styles.
• TagMatch Approach: Aimed at enhancing interpretability, TagMatch annotates artworks using a selective, multilabel classification to assign atomic tags depicting various stylistic aspects. These tags are then aggregated to formulate unique tag signatures for each artist. This mechanism not only provides a method for style classification but also contributes to an in-depth understanding of the constituents of an artist's style.

Empirical Analysis

Upon applying ArtSavant to a dataset of images generated by contemporary text-to-image models, prompted to imitate the styles of artists in the reference dataset, it was found that only about 20% of the artists' styles were at a significant risk of being copied. This outcome highlights the not yet pervasive but concerning occurrence of style mimicry by generative models. It also underscores a potential future trajectory wherein such infringement could become more widespread as generative technologies advance.

Implications and Future Directions

• Theoretical Implications: This paper provides a novel framework for understanding and analyzing the nuanced concept of artistic style in the digital era, offering a solid ground for future research in digital arts, copyright law, and AI.
• Practical Implications: For artists, legal professionals, and policymakers, the development of ArtSavant presents a practical tool for identifying and arguing cases of stylistic copyright infringement, which could inform policy adjustments and legal standards.
• Future Developments: Anticipated advancements in interpretability and attribution methodologies within AI could further refine our ability to identify and protect unique artistic styles against unauthorized replication by generative models.

Conclusion

This paper marks a significant step towards addressing copyright issues in the era of generative models. By reframing artistic style infringement as a classification problem and developing ArtSavant, it lays down a methodology for quantitatively and qualitatively analysing style copying. Future enhancements in computational methods and legal frameworks are crucial to safeguard artists against the evolving capabilities of AI-driven content generation technologies.